Armchair Architects: Multi-agent Orchestration and Patterns

Description

This episode of the Azure Essentials Show brings the Armchair Architects—Uli, Eric, and David—together to tackle multi-agent architecture in the enterprise. You'll get practical insights into orchestrating AI agents, preventing data leaks, managing costs, and choosing the right design patterns. The trio highlight real-world strategies, emerging tools, and actionable frameworks to help you build secure, scalable agent systems. Resources - Azure Architecture Center https://learn.microsoft.com/azure/architecture/ - AutoGen https://www.microsoft.com/research/project/autogen - Introduction to Semantic Kernel https://learn.microsoft.com/semantic-kernel/overview/ - Microsoft Purview https://learn.microsoft.com/purview/ - Microsoft Fabric documentation https://learn.microsoft.com/fabric/ - Agent Framework documentation https://learn.microsoft.com/agent-framework/ Three important things you will learn - How to architect and orchestrate multiple AI agents, including key patterns like specialist delegation, sequential workflows, and parallel agent reasoning. - The importance of data leak prevention, security policies, and governance when deploying agents in enterprise environments. - Practical considerations for agent design, such as cost management, model selection, and leveraging message-based systems for scalability and reliability. Recommended Next Steps - Review and apply message-based architectural patterns (such as those used in message brokers) as a foundation for agent orchestration. - Explore open-source frameworks like Microsoft Research Autogen and Semantic Kernel to understand and experiment with multi-agent patterns. - Assess and implement data governance and security controls, including data leak prevention, for agent interactions within your organization. Related Episodes - Watch more episodes of Armchair Architects https://aka.ms/ArmchairArchitects - Watch more episodes of the Azure Essentials Show https://aka.ms/AzureEssentialsShow Connect - David Blank-Edelman https://www.linkedin.com/in/dnblankedelman/ - Uli Homann https://www.linkedin.com/in/ulrichhomann/ - Eric Charran https://www.linkedin.com/in/ericcharran/ Chapters 00:00 Hey! Good news! 00:29 Welcome architects 01:17 Top things to consider 02:30 Data Leak Prevention (DLP) 04:21 Focused agents 05:45 Agent-to-agent conversations 07:04 Agents as employees 08:36 Cost considerations 09:00 Pattern considerations 11:15 Regression testing 12:12 Parallel pattern 13:38 Sequential pattern 13:51 Specialist pattern 15:05 Search engine example 15:35 Shopping agent example 15:55 Time as a factor 16:30 Negotiation and consensus pattern 17:43 Reviewer pattern 18:58 Message-based patterns 20:40 Check out AutoGen 22:07 PII Trust boundaries